Monday, July 29, 2013

Book Review: The Quants



If you want a fun, non-technical history of quantitative finance, this is your book. It traces the development of quant models, from Ed Thorp and Black-Scholes-Merton all the way through David X. Li. It explains trading strategies like statistical arbitrage in layman's terms, and offers insight into what it's like to run a trading operation, day to day. And it narrates the three-decade-long run-up to the epic finance-industry meltdown of 2007-8, including "warm-ups" like Black Monday and the fall of Long Term Capital Management. The author, Scott Patterson, writes in an accessible, engaging style, making it difficult to put the book down at times. It also does a good job of profiling some of the colorful personalities of the quant world - Peter Muller and Cliff Asness being the most colorful of the bunch.

So I definitely recommend The Quants.

Like (almost) all books, The Quants is not without its flaws. Chief among these is that the book mostly fails to deliver on one of the promises in its subtitle - it fails to give a detailed account of the quants that nearly destroyed Wall Street.

Why was Wall Street almost destroyed? The general story that I buy goes something like this:

1) Wall Street firms, by creating securitized mortgage-backed financial products, increased the demand for housing, which along with other factors helped fuel a nationwide housing bubble.

(Note: This assumes that some initial rapid price increase is needed to "jump-start" a bubble, leading to a spiral of self-reinforcing price increases, either through speculation, herd behavior, or the mistaken inference of underlying structural change, or some combination thereof. That may not be how bubbles really work, but it's currently my best guess, from what I've seen in the lab and read in the literature.)

2) Wall Street firms priced these securities using quantitative models that were fundamentally flawed, leading them to understate the risk of the mortgage-backed securities and derivatives based on these securities. Some of the flaws included eternally rising national house prices, non-time-varying correlations, thin-tailed (less risky) stochastic processes, and omission of counterparty risk.

3) These models lead banks to hold way too many of the aforementioned securities, and to borrow way too much money (leverage) to buy them. Complacent (and possibly captured) regulators and ratings agencies did nothing to stop this.

4) When the bubble popped and prices fell and correlations spiked, banks collapsed.

This story is very similar to the one given in The Quants. So Patterson does, in my view, get the crisis correct.

But four of the six people profiled in The Quants were operators of independent hedge funds - Ed Thorp of Princeton Newport, Ken Griffin of Citadel, Cliff Asness of AQR, and Jim Simons of Renaissance. Though big by historical hedge fund standards, these firms were not nearly as big (or, for the most part, as leveraged) as the banks that were at the center of the crisis - Goldman Sachs, Bank of America, Lehman, etc. And though most of the quant hedge funds suffered in the great 2008 crash, the specter of their collapse did not really threaten the system itself. In fact, as The Quants recounts, the quant hedge funds suffered their own meltdown a year earlier, in the summer of 2007, but this caused far fewer problems for the economy than the near-collapse of the big banks in 2008. Notably, none of the independent quant hedge funds were bailed out by the U.S. government in 2008 - some because they were not in danger, others because they simply weren't systematically important enough to warrant bailouts.

This is not to say that quant hedge funds can't be dangerous to the financial system; Long Term Capital Management clearly was. It's simply the case that this time, in 2008, independent hedge funds were not at the center of the crisis. They mostly sat on the sidelines, watching the carnage and hoping that they didn't die. One of the funds Patterson profiles, Renaissance Technologies, was using strategies so different from the big banks that it gained 80% in 2008. (Disclosure: The founder and some ex-employees of Renaissance Technologies have made large donations to the university where I work!)

The Quants also profiled two men who headed internal hedge funds within big banks - Peter Muller of Morgan Stanley, and Boaz Weinstein of Deutsche Bank. Weinstein's group appears to have bought lots of mortgage-backed securities, and to have lost a lot of money for the firm in the 2008 crash. But this means that only one out of the six protagonists of The Quants could plausibly be described as having "almost destroyed Wall Street".

Personally, I would have liked to have seen The Quants tell the stories of the "pricing quants" and "risk quants" within the big banks, whose models were instrumental in convincing regulators, ratings agencies, and bank executives themselves that mortgage-backed products were safe. Patterson does briefly tell the story of two such quants, Fischer Black  David X. Li, but I'd like to have seen a lot more about these guys, and about academics like Myron Scholes and Robert Merton.

Did the quants within big banks know that their models were wrong? Did they try to warn the executives not to apply the models? Did they simply shut up and take a paycheck as greedy, reckless executives misused and over-applied their models? Or did they actively promote blanket and widespread use of the flawed models, encouraging the heavy use of leverage and the holding of huge amounts of mortgage-backed products? I wish The Quants had answered this question, but it did not.

I also think The Quants displays a little too much fear of the unknown. Patterson definitely seems to regard the black-box strategies of companies like Renaissance as creepy and inherently suspicious. But I think this fear is mostly unwarranted. Any assumptions can be wrong, so any algorithm can blow up. No quant model is riskless or foolproof, and none will ever be. But that applies just as much to the intuitive trades of "shoot-from-the-hip" human traders, or to any other investing mechanism. No spooky supercomputers or secretive mathematicians were needed in order to blow up the Japanese financial system in 1989, or the U.S. financial system in 1929.

The key, as Ed Thorp is quoted as saying in the book, is in limiting leverage, so that collapses don't bring down the whole system. Limiting the systemic risk caused by "too big to fail" institutions also seems to be important. The mundane factors of leverage and bigness seem much more scary, to me, than the opacity of quant trading strategies.

Nevertheless, it is a really good book, and if you have any interest in the history of finance, you should check it out.

27 comments:

  1. Anonymous4:54 AM

    Once again, you (and the authors) get the housing bubble wrong.

    Tell us again how Wall Street created a housing bubble in Spain and now Toronto. This ought to be a good laugh.

    Listen, I hate Wall Street as much as the next guy. But it's important to accurately describe what happened if economists are to be taken seriously.

    ReplyDelete
    Replies
    1. Anonymous4:59 AM

      Further reading: http://www.guardian.co.uk/commentisfree/cifamerica/2010/mar/08/financial-crisis-subprimecrisis

      Delete
    2. You're right, I should have specified that securitization was probably not the only factor that created the housing bubble. I will edit.

      Delete
    3. I should also mention that I don't hate finance. I just think it's often dangerous, and sometimes wasteful, and we need to make it function better.

      Delete
    4. The housing bubble probably started as a reaction to the stock market crash in 2001 and low interest rates resulting in a shift towards investing in housing. What enabled the bubble to grow as it did was that banks kept feeding money into the fire. German banks were pouring money into both the American and the Spanish mortgage systems so Spain is hardly a counter-example.

      Canada had and continues to have high prices in a number of markets but that may be a result of stable banking systems, low interest rates, oil and natural resources and foreign investment in real estate in major centers (Vancouver and Toronto).

      Delete
    5. Anonymous7:24 PM

      Banks in those countries created their own housing bubbles using the same techniques, no?

      Delete
  2. Spain's housing market was financed by capital flows from abroad (e.g. Germnay). So the the statement that the system was "well regulated" needs to be modified since it did not take into account that systemic risk.

    However, it is also probably true that the real failure was that globalization of the capital markets means that one idea tends to infect all players, much as an agricultural monoculture makes crops susceptible to massive wipeout with single pests. The idea was that debt in general was less risky than it actually was. Not just real estate but all types of loans. Both lenders and creditors thought this. Minsky wins again.

    ReplyDelete
  3. Anonymous7:11 AM

    Regarding the views of users of quantitative risk management, Donald MacKenzie has some interesting work, based on interviews, about the use of Gaussian copulas. See http://www.socialwork.ed.ac.uk/__data/assets/pdf_file/0003/84243/Gaussian14.pdf

    The way he describes it, everyone at least claims to be aware of the limitations of the formulas and insists that they don't simply follow the model as is, but they use it for communication within the firm, with counterparties, and with clients (easier to quote a few model parameters than a general impression), and, importantly, for accounting purposes. Perhaps unsurprisingly given the author, analogy is made with Black Scholes, where options are still quoted based on 'implied volatility', even though no option trader uses Black Scholes as their quantitative model of options for analysis and trading purposes anymore.

    Still, I agree that a sense of what was going on intellectually within the investment banks and other financial companies which played major roles in the crisis deserves to be better explored, though of course everyone involved has incentive to point fingers elsewhere.

    ReplyDelete
    Replies
    1. Anonymous11:24 AM

      The Gaussian distribution is the one and ONLY distribution that arises out of the Black-Scholes no-arbitrage PDE. I keep reading people criticize the distribution as being inaccurate, and maybe it is, but if we choose another distribution then the assumption changes. Its not like the creators of BS chose this distribution arbitrary.

      Delete
  4. From your description, Noah, I think 'The Big Short' has more about CDO traders within the banks themselves. Have you read it?

    With regards to Wall Street and Spain: There might be no direct links but, as someone else pointed out, it was foreign capital that helped keep prices buoyant. In Greece, French and German capital was heavily invested. Ditto Eastern Europe...

    Speculation is as speculation does...

    ReplyDelete
    Replies
    1. Yep I read it, and it is awesome, but it's also mostly from the perspective of outsiders. I want the inside dope on what quants were doing inside BofA.

      Delete
  5. Like Frederic said, The Big Short calls it well, especially Dr. Michael Burry, who read the actual contracts about MBS and CDOs before designing the Short which would make his Scion Fund billions when the clear-to-him housing bubble popped.

    LTCM was clearly dangerous << sorry, I don't know all details, but I think the "easy" big gov't bailout to avoid a messy bankruptcy was, in fact, more dangerous. It established the precedent of Too Big To Fail, rather than, bankruptcy can come to anybody.

    Moral Hazard is real and LTCM pushed speculators to be big enough to qualify. Allowing Lehman to go under was partly in recognition that the TBTF policy is not good -- but then not allowing AIG (and thus Goldman) to go belly up shows unjust capriciousness.

    ReplyDelete
  6. Diego Espinosa9:59 AM

    The problem with quants is that they can exhibit homogeneity, and their models do not capture the effect this has on systemic risk. They therefore increase fragility and the probability of catastrophic loss.

    ReplyDelete
  7. But the point that was made was that this was not just a Wall Street phenomenon. So while US policy was poor, the issue is whether that was even necessary for the bubble to have arisen. More likely it was mutual delusional thinking on part of both regulators and regulated.

    More interestingly, there was the clear idea that MBS and other consumer type loans had a better risk/return ratio than alternative investments (corporate bonds, equity, etc). Perhaps this goes along with creation of any new asset class, as MBS seemed to be at the time. Has anyone studied that?


    ReplyDelete
  8. Their models were not wrong, any more than IS/LM or the Ricardian equivalence were wrong. They just didn't apply in the then-current situation. When the facts change, one must change the model you are curve-fitting the data to.

    According to Black-Scholes, my Solyndra options are still intrinsically worth $50 million, btw.

    ReplyDelete
  9. o. nate10:48 AM

    IMO, the strongest case to be made that quants nearly destroyed Wall Street is all about one asset class: ABS CDOs backed by subprime RMBS. This one asset class was responsible for the biggest losses at the biggest banks. It was also the most severely mis-rated asset class by the rating agencies: AFAIK, it's the only product in which AAA-rated assets suffered significant losses. How did the models screw this one up so badly? Basically it comes down to correlation. It turns out that in the event of a nationwide decline in housing prices, losses on subprime RMBS mezzanine tranches (the underlying assets of the CDO) are quite highly correlated. The models that were used to rate these tranches were fairly sophisticated, but they depend on having good data, and the historical data that was used did not include any period of nationwide housing decline. Someone should have noticed that - it remains somewhat of a mystery that so few did.

    ReplyDelete
  10. Noah,

    Great piece, and very accurate in my opinion. I would add there were two roles significant relevant quant roles within investment banks.

    The first was in developing the models that front office used to value and hedge their books and also the models (sometimes the same) used by risk management. In this area there was a huge discrepancy between different dealers, with the best run (JP, GS) managing impressively well under the circumstances. I don't know what models were in use at Bear and Lehman (not much from what I've heard), but I am quite certain that GS was not running their trading book off the Gaussian copula. The Gaussian copula, and related models, is simply inappropriate for dynamic hedging because it is a single period default model (the greeks are not time-consistent and the model cannot incorporate spread dynamics). Why some dealers managed their own books better than others is simply cultural. To the extent that Lehman may have used the Gaussian copula or other stupid models in managing their books, I would blame the power of traders and management in creating a dangerous risk management environment.

    The second role of quants was in engineering the structured credit trades that were foisted on the world, especially in the period from about 2004-2008. These sell side quants worked actively and closely with quants at the rating agencies on each structured deal with the principal goal of producing the most worthless possible product with a given credit rating. Making this trick work well depended critically on the rating agency models being as crappy and unrealistic as possible, and the agencies cooperated by 1) never changing model type (always the Gaussian copula for CDO's, CDO^2, CMBS, RMBS, etc) and, 2) even when the underlying credits were obviously extremely correlated (mezz subprime RMBS tranches), using absurdly low correlation assumptions. Unlike the first role above, this quant role was absolutely instrumental in driving the bubble.

    Blaming the quants, of course, is as useful as blaming the other bankers, i.e. totally pointless. Bankers and quants didn't suddenly turn evil any more than borrowers suddenly turned "predatory." I actually do blame government, though nothing to do with the GSEs or lack of regulation.

    The real problem is excessive bad regulation, much of which has to do with the rating agencies. Bank and insurance company regulatory capital calculations and eligible money market instrument rules are, to a large extent, determined by the ratings provided by the government approved rating agencies (NRSROs). The rules give the agencies privileged information about the rated instruments, and in the case of money market instruments it is often the case that almost no information is available to the investor other than the rating. This, in turn, incentivizes the money market (or insurance co, pension fund, etc.) fund manager to simply buy whatever is cheapest at the given rating, with essentially no accountability. The NRSROs, of course, are paid by the issuers, and the courts have determined that their right to be catastrophically wrong is constitutionally protected by the first amendment.

    A big part of the solution is to simple get rid of government sanctioned and mandated investment advice, and thereby require investors to perform the role that efficient market theory expects them to perform. We have no use for the NRSROs and, on the contrary, they have proven themselves to be totally incompetent and corrupt.

    ReplyDelete
    Replies
    1. Yep, I want to hear the stories of the risk quants and the pricing quants and the sell-side quants. I'm not as confident as you are that the internal pricing quants mostly avoided reliance on stupid models for the valuation of banks' portfolios...

      Delete
  11. Anonymous11:20 AM

    Let's hope Larry Summers becomes the nominee. Stocks will surge on this news because his competence is unmatched by none other, except for perhaps Bernanke.

    Quant Finance is useful and wasn't the cause of crisis, but even if it were it's easy to fix the crisis and make it go away by recapitalization. A crisis drives rates lower making the bailout painless.

    ReplyDelete
  12. To fix finance we just need a quant model that estimates the riskiness of quant models. Makes sense.

    ReplyDelete
  13. I see that while I was writing o. nate made the same comment about mezz rmbs tranches. I agree that, as esoteric as it sounds, it was a fairly critical error in enabling an absolutely huge proportion of the nastiest issuances.

    For all the faults of the Gaussian copula, it should be said that from a *realized* default perspective it didn't turn out that bad if you are looking at corporate defaults. Mezz and senior tranches of corporate CDO's, like the ones which were the bread and butter of the most responsible dealers (JP and MS) for the most part suffered no or small defaults.

    The Gaussian copula (GC) has flaws that can be arbitraged when structuring CDOs containing other CDOs (CDO-squareds). To their credit, however, the rating agencies at least actually used the GC in the theoretically correct way (looking through the embedded CDOs) when evaluating corporate CDO^2s. However, for some reason it became standard practice when evaluating CDOs containing residential mortgage backed bonds (RMBS), not to look through to the underlying mortgages, and instead treat the RMBS as simple securities, even though they were, in fact, mezzanine tranches of mortgage pools. Now, any idiot can see that if there is a large number of mortgage defaults, then the defaults of mezzanine RMBS tranches will be a) widespread and b) almost perfectly correlated. The agencies just stuck arbitrary, low correlations on the RMBSs (0.2 or so, IIRC?).

    I don't know when this particular "mistake" started but I am aware of it going on at least back to 2004. Because the error was so huge it enabled absolute garbage to be used as backing for AAA commercial paper in a way that could never have been done on an economically efficient balance sheet, and led to massive growth of RMBS/CMBS SIVs, ABCP, etc.

    ReplyDelete
  14. I think point 3 of your summary needs just a touch more nuance, accounting for the profitability of the fees generated by turning out more securitized products as another factor that led to banks holding too many of them.

    ReplyDelete
  15. Just got a copy, noah

    ReplyDelete
    Replies
    1. Actually Sophia made me read it in the first place...

      Delete
  16. Basically the book ignores the issues related to the financial crisis. It is mostly about hedge funds and not about securitization. It is about big hedge fund traders who were mostly orthogonal to the financial crisis. So I don't understand why you like the book. I read it, agreed with much of what you wrote, so I did not like it, for the reasons you outline. It does not even understand quantitative finance very well.

    ReplyDelete
  17. Anonymous4:47 PM

    (Ragweed, actually)

    Noah - have your read Nick Dunbar on the crisis? He was not an insider, but as a Journalist for insider publications like Risk he had a lot of access. And he started out in physics too.

    ReplyDelete
  18. Noah,

    The problem with the 'quants fooled the innocent executives' theory is that it's really unlikely. If somebody's been on Wall St for thirty years, they've (a) seen massive changes and turmoil in the markets, and (b) have seen a thousand bright young math-heavy guys with schemes for free money. Considering the fact that the higher-ups made massive sums of money through these frauds, the more likely story is that they wanted it that way, and in general the quants were serving the higher ups.

    ReplyDelete